With 3.8-rc1, the first call of pci_map_single() that is not checked
with a corresponding pci_dma_mapping_error() call results in a warning
with a splat as follows:
WARNING: at lib/dma-debug.c:933 check_unmap+0x480/0x950()
Hardware name: HP Pavilion dv2700 Notebook PC
forcedeth 0000:00:0a.0: DMA-API: device driver failed to check
map error[device address=0x00000000b176e002] [size=90 bytes] [mapped as single]
Signed-off-by: Larry Finger <[email protected]>
---
drivers/net/ethernet/nvidia/forcedeth.c | 23 +++++++++++++++++++++++
1 file changed, 23 insertions(+)
diff --git a/drivers/net/ethernet/nvidia/forcedeth.c b/drivers/net/ethernet/nvidia/forcedeth.c
index 653487d..de39cf2 100644
--- a/drivers/net/ethernet/nvidia/forcedeth.c
+++ b/drivers/net/ethernet/nvidia/forcedeth.c
@@ -1821,6 +1821,11 @@ static int nv_alloc_rx(struct net_device *dev)
skb->data,
skb_tailroom(skb),
PCI_DMA_FROMDEVICE);
+ if (pci_dma_mapping_error(np->pci_dev,
+ np->put_rx_ctx->dma)) {
+ dev_kfree_skb_any(skb);
+ goto packet_dropped;
+ }
np->put_rx_ctx->dma_len = skb_tailroom(skb);
np->put_rx.orig->buf = cpu_to_le32(np->put_rx_ctx->dma);
wmb();
@@ -1830,6 +1835,7 @@ static int nv_alloc_rx(struct net_device *dev)
if (unlikely(np->put_rx_ctx++ == np->last_rx_ctx))
np->put_rx_ctx = np->first_rx_ctx;
} else {
+packet_dropped:
u64_stats_update_begin(&np->swstats_rx_syncp);
np->stat_rx_dropped++;
u64_stats_update_end(&np->swstats_rx_syncp);
@@ -1856,6 +1862,11 @@ static int nv_alloc_rx_optimized(struct net_device *dev)
skb->data,
skb_tailroom(skb),
PCI_DMA_FROMDEVICE);
+ if (pci_dma_mapping_error(np->pci_dev,
+ np->put_rx_ctx->dma)) {
+ dev_kfree_skb_any(skb);
+ goto packet_dropped;
+ }
np->put_rx_ctx->dma_len = skb_tailroom(skb);
np->put_rx.ex->bufhigh = cpu_to_le32(dma_high(np->put_rx_ctx->dma));
np->put_rx.ex->buflow = cpu_to_le32(dma_low(np->put_rx_ctx->dma));
@@ -1866,6 +1877,7 @@ static int nv_alloc_rx_optimized(struct net_device *dev)
if (unlikely(np->put_rx_ctx++ == np->last_rx_ctx))
np->put_rx_ctx = np->first_rx_ctx;
} else {
+packet_dropped:
u64_stats_update_begin(&np->swstats_rx_syncp);
np->stat_rx_dropped++;
u64_stats_update_end(&np->swstats_rx_syncp);
@@ -2217,6 +2229,9 @@ static netdev_tx_t nv_start_xmit(struct sk_buff *skb, struct net_device *dev)
bcnt = (size > NV_TX2_TSO_MAX_SIZE) ? NV_TX2_TSO_MAX_SIZE : size;
np->put_tx_ctx->dma = pci_map_single(np->pci_dev, skb->data + offset, bcnt,
PCI_DMA_TODEVICE);
+ if (pci_dma_mapping_error(np->pci_dev,
+ np->put_tx_ctx->dma))
+ return NETDEV_TX_BUSY;
np->put_tx_ctx->dma_len = bcnt;
np->put_tx_ctx->dma_single = 1;
put_tx->buf = cpu_to_le32(np->put_tx_ctx->dma);
@@ -2337,6 +2352,9 @@ static netdev_tx_t nv_start_xmit_optimized(struct sk_buff *skb,
bcnt = (size > NV_TX2_TSO_MAX_SIZE) ? NV_TX2_TSO_MAX_SIZE : size;
np->put_tx_ctx->dma = pci_map_single(np->pci_dev, skb->data + offset, bcnt,
PCI_DMA_TODEVICE);
+ if (pci_dma_mapping_error(np->pci_dev,
+ np->put_tx_ctx->dma))
+ return NETDEV_TX_BUSY;
np->put_tx_ctx->dma_len = bcnt;
np->put_tx_ctx->dma_single = 1;
put_tx->bufhigh = cpu_to_le32(dma_high(np->put_tx_ctx->dma));
@@ -5003,6 +5021,11 @@ static int nv_loopback_test(struct net_device *dev)
test_dma_addr = pci_map_single(np->pci_dev, tx_skb->data,
skb_tailroom(tx_skb),
PCI_DMA_FROMDEVICE);
+ if (pci_dma_mapping_error(np->pci_dev,
+ test_dma_addr)) {
+ dev_kfree_skb_any(tx_skb);
+ goto out;
+ }
pkt_data = skb_put(tx_skb, pkt_len);
for (i = 0; i < pkt_len; i++)
pkt_data[i] = (u8)(i & 0xff);
--
1.7.10.4
On Thu, 2012-12-27 at 14:38 -0600, Larry Finger wrote:
> On 12/27/2012 02:05 PM, Eric Dumazet wrote:
> > On Thu, 2012-12-27 at 13:42 -0600, Larry Finger wrote:
> >> + if (pci_dma_mapping_error(np->pci_dev,
> >> + np->put_tx_ctx->dma))
> >> + return NETDEV_TX_BUSY;
> >
> > Really this is not going to work very well : caller will call this in a
> > loop.
>
> Any suggestions on what value should be returned, or does the caller need to be
> modified?
NETDEV_TX_BUSY is really obsolete
Documentation/networking/driver.txt
In case of mapping error, I would drop the packet.
(kfree_skb() it, increment a device tx_dropped counter, and return
NETDEV_TX_OK)
From: Eric Dumazet <[email protected]>
Date: Thu, 27 Dec 2012 13:03:29 -0800
> On Thu, 2012-12-27 at 14:38 -0600, Larry Finger wrote:
>> On 12/27/2012 02:05 PM, Eric Dumazet wrote:
>> > On Thu, 2012-12-27 at 13:42 -0600, Larry Finger wrote:
>
>> >> + if (pci_dma_mapping_error(np->pci_dev,
>> >> + np->put_tx_ctx->dma))
>> >> + return NETDEV_TX_BUSY;
>> >
>> > Really this is not going to work very well : caller will call this in a
>> > loop.
>>
>> Any suggestions on what value should be returned, or does the caller need to be
>> modified?
>
> NETDEV_TX_BUSY is really obsolete
>
> Documentation/networking/driver.txt
>
> In case of mapping error, I would drop the packet.
Agreed.
On 12/27/2012 02:05 PM, Eric Dumazet wrote:
> On Thu, 2012-12-27 at 13:42 -0600, Larry Finger wrote:
>> With 3.8-rc1, the first call of pci_map_single() that is not checked
>> with a corresponding pci_dma_mapping_error() call results in a warning
>> with a splat as follows:
>>
>> WARNING: at lib/dma-debug.c:933 check_unmap+0x480/0x950()
>> Hardware name: HP Pavilion dv2700 Notebook PC
>> forcedeth 0000:00:0a.0: DMA-API: device driver failed to check
>> map error[device address=0x00000000b176e002] [size=90 bytes] [mapped as single]
>>
>> Signed-off-by: Larry Finger <[email protected]>
>> ---
>> drivers/net/ethernet/nvidia/forcedeth.c | 23 +++++++++++++++++++++++
>> 1 file changed, 23 insertions(+)
>>
>> diff --git a/drivers/net/ethernet/nvidia/forcedeth.c b/drivers/net/ethernet/nvidia/forcedeth.c
>> index 653487d..de39cf2 100644
>> --- a/drivers/net/ethernet/nvidia/forcedeth.c
>> +++ b/drivers/net/ethernet/nvidia/forcedeth.c
>> @@ -1821,6 +1821,11 @@ static int nv_alloc_rx(struct net_device *dev)
>> skb->data,
>> skb_tailroom(skb),
>> PCI_DMA_FROMDEVICE);
>> + if (pci_dma_mapping_error(np->pci_dev,
>> + np->put_rx_ctx->dma)) {
>> + dev_kfree_skb_any(skb);
>
> skb has no destructor yet, kfree_skb(skb) should be fine
OK.
>
>> + goto packet_dropped;
>> + }
>> np->put_rx_ctx->dma_len = skb_tailroom(skb);
>> np->put_rx.orig->buf = cpu_to_le32(np->put_rx_ctx->dma);
>> wmb();
>> @@ -1830,6 +1835,7 @@ static int nv_alloc_rx(struct net_device *dev)
>> if (unlikely(np->put_rx_ctx++ == np->last_rx_ctx))
>> np->put_rx_ctx = np->first_rx_ctx;
>> } else {
>> +packet_dropped:
>> u64_stats_update_begin(&np->swstats_rx_syncp);
>> np->stat_rx_dropped++;
>> u64_stats_update_end(&np->swstats_rx_syncp);
>> @@ -1856,6 +1862,11 @@ static int nv_alloc_rx_optimized(struct net_device *dev)
>> skb->data,
>> skb_tailroom(skb),
>> PCI_DMA_FROMDEVICE);
>> + if (pci_dma_mapping_error(np->pci_dev,
>> + np->put_rx_ctx->dma)) {
>> + dev_kfree_skb_any(skb);
>> + goto packet_dropped;
>> + }
>> np->put_rx_ctx->dma_len = skb_tailroom(skb);
>> np->put_rx.ex->bufhigh = cpu_to_le32(dma_high(np->put_rx_ctx->dma));
>> np->put_rx.ex->buflow = cpu_to_le32(dma_low(np->put_rx_ctx->dma));
>> @@ -1866,6 +1877,7 @@ static int nv_alloc_rx_optimized(struct net_device *dev)
>> if (unlikely(np->put_rx_ctx++ == np->last_rx_ctx))
>> np->put_rx_ctx = np->first_rx_ctx;
>> } else {
>> +packet_dropped:
>> u64_stats_update_begin(&np->swstats_rx_syncp);
>> np->stat_rx_dropped++;
>> u64_stats_update_end(&np->swstats_rx_syncp);
>> @@ -2217,6 +2229,9 @@ static netdev_tx_t nv_start_xmit(struct sk_buff *skb, struct net_device *dev)
>> bcnt = (size > NV_TX2_TSO_MAX_SIZE) ? NV_TX2_TSO_MAX_SIZE : size;
>> np->put_tx_ctx->dma = pci_map_single(np->pci_dev, skb->data + offset, bcnt,
>> PCI_DMA_TODEVICE);
>> + if (pci_dma_mapping_error(np->pci_dev,
>> + np->put_tx_ctx->dma))
>> + return NETDEV_TX_BUSY;
>
> Really this is not going to work very well : caller will call this in a
> loop.
Any suggestions on what value should be returned, or does the caller need to be
modified?
Thanks for the review,
Larry
On Thu, 2012-12-27 at 13:42 -0600, Larry Finger wrote:
> With 3.8-rc1, the first call of pci_map_single() that is not checked
> with a corresponding pci_dma_mapping_error() call results in a warning
> with a splat as follows:
>
> WARNING: at lib/dma-debug.c:933 check_unmap+0x480/0x950()
> Hardware name: HP Pavilion dv2700 Notebook PC
> forcedeth 0000:00:0a.0: DMA-API: device driver failed to check
> map error[device address=0x00000000b176e002] [size=90 bytes] [mapped as single]
>
> Signed-off-by: Larry Finger <[email protected]>
> ---
> drivers/net/ethernet/nvidia/forcedeth.c | 23 +++++++++++++++++++++++
> 1 file changed, 23 insertions(+)
>
> diff --git a/drivers/net/ethernet/nvidia/forcedeth.c b/drivers/net/ethernet/nvidia/forcedeth.c
> index 653487d..de39cf2 100644
> --- a/drivers/net/ethernet/nvidia/forcedeth.c
> +++ b/drivers/net/ethernet/nvidia/forcedeth.c
> @@ -1821,6 +1821,11 @@ static int nv_alloc_rx(struct net_device *dev)
> skb->data,
> skb_tailroom(skb),
> PCI_DMA_FROMDEVICE);
> + if (pci_dma_mapping_error(np->pci_dev,
> + np->put_rx_ctx->dma)) {
> + dev_kfree_skb_any(skb);
skb has no destructor yet, kfree_skb(skb) should be fine
> + goto packet_dropped;
> + }
> np->put_rx_ctx->dma_len = skb_tailroom(skb);
> np->put_rx.orig->buf = cpu_to_le32(np->put_rx_ctx->dma);
> wmb();
> @@ -1830,6 +1835,7 @@ static int nv_alloc_rx(struct net_device *dev)
> if (unlikely(np->put_rx_ctx++ == np->last_rx_ctx))
> np->put_rx_ctx = np->first_rx_ctx;
> } else {
> +packet_dropped:
> u64_stats_update_begin(&np->swstats_rx_syncp);
> np->stat_rx_dropped++;
> u64_stats_update_end(&np->swstats_rx_syncp);
> @@ -1856,6 +1862,11 @@ static int nv_alloc_rx_optimized(struct net_device *dev)
> skb->data,
> skb_tailroom(skb),
> PCI_DMA_FROMDEVICE);
> + if (pci_dma_mapping_error(np->pci_dev,
> + np->put_rx_ctx->dma)) {
> + dev_kfree_skb_any(skb);
> + goto packet_dropped;
> + }
> np->put_rx_ctx->dma_len = skb_tailroom(skb);
> np->put_rx.ex->bufhigh = cpu_to_le32(dma_high(np->put_rx_ctx->dma));
> np->put_rx.ex->buflow = cpu_to_le32(dma_low(np->put_rx_ctx->dma));
> @@ -1866,6 +1877,7 @@ static int nv_alloc_rx_optimized(struct net_device *dev)
> if (unlikely(np->put_rx_ctx++ == np->last_rx_ctx))
> np->put_rx_ctx = np->first_rx_ctx;
> } else {
> +packet_dropped:
> u64_stats_update_begin(&np->swstats_rx_syncp);
> np->stat_rx_dropped++;
> u64_stats_update_end(&np->swstats_rx_syncp);
> @@ -2217,6 +2229,9 @@ static netdev_tx_t nv_start_xmit(struct sk_buff *skb, struct net_device *dev)
> bcnt = (size > NV_TX2_TSO_MAX_SIZE) ? NV_TX2_TSO_MAX_SIZE : size;
> np->put_tx_ctx->dma = pci_map_single(np->pci_dev, skb->data + offset, bcnt,
> PCI_DMA_TODEVICE);
> + if (pci_dma_mapping_error(np->pci_dev,
> + np->put_tx_ctx->dma))
> + return NETDEV_TX_BUSY;
Really this is not going to work very well : caller will call this in a
loop.
> np->put_tx_ctx->dma_len = bcnt;
> np->put_tx_ctx->dma_single = 1;
> put_tx->buf = cpu_to_le32(np->put_tx_ctx->dma);
> @@ -2337,6 +2352,9 @@ static netdev_tx_t nv_start_xmit_optimized(struct sk_buff *skb,
> bcnt = (size > NV_TX2_TSO_MAX_SIZE) ? NV_TX2_TSO_MAX_SIZE : size;
> np->put_tx_ctx->dma = pci_map_single(np->pci_dev, skb->data + offset, bcnt,
> PCI_DMA_TODEVICE);
> + if (pci_dma_mapping_error(np->pci_dev,
> + np->put_tx_ctx->dma))
> + return NETDEV_TX_BUSY;
same problem here.
> np->put_tx_ctx->dma_len = bcnt;
> np->put_tx_ctx->dma_single = 1;
> put_tx->bufhigh = cpu_to_le32(dma_high(np->put_tx_ctx->dma));
> @@ -5003,6 +5021,11 @@ static int nv_loopback_test(struct net_device *dev)
> test_dma_addr = pci_map_single(np->pci_dev, tx_skb->data,
> skb_tailroom(tx_skb),
> PCI_DMA_FROMDEVICE);
> + if (pci_dma_mapping_error(np->pci_dev,
> + test_dma_addr)) {
> + dev_kfree_skb_any(tx_skb);
kfree_skb(skb);
> + goto out;
> + }
> pkt_data = skb_put(tx_skb, pkt_len);
> for (i = 0; i < pkt_len; i++)
> pkt_data[i] = (u8)(i & 0xff);